Compare several survival learners on shared resampling folds

Evaluates each learner in models on the same set of resampling splits, so the per-fold values are paired across models. Returns a leaderboard plus the full per-split table.

Usage

compare(
  formula,
  data,
  models,
  specs = list(),
  times = NULL,
  resampling = cv(),
  metrics = c("cindex", "ibs"),
  tune = FALSE,
  ncores = 1,
  ...
)

Arguments

formula: A survival formula.
data: A data frame.
models: Character vector of learner ids, or a list mixing ids and custom learners (see fit()). List names, when given, label the learners.
specs: Named list of engine-argument lists, keyed by learner label (specs[["ranger"]] <- list(num.trees = 500)).
times: Evaluation grid; NULL picks up to 50 event-time quantiles (5th to 95th percentile).
resampling: A resampling spec (default cv()); the split set is drawn once and reused for every model.
metrics: Character vector from listmetrics()$metric.
tune: Logical. When TRUE, each tunable model is tuned once on the full data (its default grid) and the winning configuration is then evaluated on the shared folds. This is not nested CV, so the tuned results are optimistic; a message is emitted. Default FALSE.
ncores: Cores for the split loop (default 1).
...: Unused.

Value

A survalis_compare object: leaderboard (model x metric, mean and se), per_split (long), times, resampling, metrics, models, tuned.

Examples

cmp <- compare(Surv(time, status) ~ age + karno + celltype, veteran,
               models = c("coxph", "ranger"), times = c(90, 180, 365),
               resampling = cv(3))
cmp
#> <survalis_compare> 2 models
#> <survalis_resampling> 3-fold CV, strata = status
#> 
#>   cindex (higher is better):
#>    * coxph            0.7282  (se 0.0229)
#>      ranger           0.6617  (se 0.0300)
#> 
#>   ibs (lower is better):
#>    * coxph            0.1496  (se 0.0071)
#>      ranger           0.1609  (se 0.0046)
plot(cmp)